Analysis And Assessment Of Android Virtual Machine Runtime Technologies

As a mainstream operating system for mobile platforms, Android reached 76.6% share of the smartphone market by the end of 2014. To support Java application for different hardware platforms, Google designed Dalvik runtime as the first Android virtual machine. However, with the improvement of the mobile platform hardware, Dalvik Just In Time compiler cannot meet the needs of today’s hardware, as a result, a new runtime named ART has been introduced in Android 4.4, which uses Ahead Of Time compiler to improve the performance of Java programs. Studying and evaluating these two runtimes are meaningful for optimizing Android system.Through deep analysis of the compiler system and running process of the two runtimes, this thesis explores how does ART improve the performance by compile bytecode to native code ahead of time. This thesis evaluates the two runtimes from three aspects:First, tests system booting and application installation performance using Bootchart and Traceview, and proposes a method to improve the booting speed by compiling the Java libs on a host computer; Second, computing performance is assessed by CaffeineMark, SciMark2.0 and BenchmarkPi, and experiments are designed to analysis the main different items; At last, comprehensive system performance is tested by OxBenchmark and Quadrant Standard, which includes graph processing, memory management and storage performance as well.The experiments results show that, the average applications installation time of ART is 1.73 times of Dalvik’s, and the file size ART generated, on average, is 2 times bigger than Dalvik’s. The system’s first booting time of ART is 2.6 times of Dalvik’s, while after using the optimation method proposed in this thesis, the booting time of ART reduces to 1.078 times of Dalvik’s. Compared to Dalvik, ART’s CaffeineMark computing performance score improves about 67%, and BenchmarkPi computing speed improves about 17.8%. In the test of comprehensive system performance, the garbage collection consuming time of ART decreases about 89% than Dalvik’s.